How can AI reduce carbon emissions? Insights from a quasi-natural experiment using generalized random forest

Abstract

This study examines the impact of a recent regional artificial intelligence pilot zone (AIPZ) policy in China on firms' carbon performance using a quasi-natural experiment. Using the Difference-in-Differences (DID) methodology, the findings reveal that the AIPZ policy significantly reduces firms' carbon emissions. This effect is most pronounced for firms with high talent levels, positive media sentiment, and strong internal control, while heavily polluting firms experience a relatively minor effect. A variable importance analysis using the generalized random forest approach identifies return on assets (ROA) and Tobin's Q as significant contributors to the variation in firms' responses. Specifically, when ROA is negative, the treatment effect is relatively large and increases slowly. In contrast, when ROA is positive, the treatment effect decreases rapidly, showing a zero-boundary effect. Additionally, Tobin's Q exhibits an inverted U-shaped relationship with the treatment effect. The findings of this study offer valuable insights for policymakers in China and beyond, highlighting the importance of considering firm-specific characteristics to achieve effective and sustainable environmental management alongside economic development.